IFRS 9 Expected Loss: A Model Proposal for Estimating the

Revista de Contabilidad Spanish Accounting Review 23 (2) (2020) 180-196

REVISTA DE CONTABILIDAD

SPANISH ACCOUNTING REVIEW

revistas.um.es/rcsar

IFRS 9 Expected Loss: A Model Proposal for Estimating the Probability ofDefault for non-rated companies

David Delgado-Vaqueroa, José Morales-Díazb, Constancio Zamora-Ramírezc

a) EY Spain. Market Risk Team (Financial Services Office). Raimundo Fernández Villaverde 65, Madrid (Spain).b) Complutense University of Madrid. Financial Administration and Accounting Department. School of Business and Economics. Campus de Somosaguas s/n, Pozuelo deAlarcón (Madrid) (Spain).c) Seville University. Accounting and Finance Department. School of Business and Economics. Ramon y Cajal 1, Seville (Spain).

aCorresponding author.E-mail address: [email protected]

A R T I C L E I N F O

Article history:Received 2 April 2019Accepted 27 June 2019Available online 1 July 2020

JEL classification:C13G33C63M41

Keywords:IFRS 9Impairment of Financial AssetsProbability of DefaultCredit Rating

A B S T R A C T

Under the IFRS 9 impairment model, entities must estimate the PD (Probability of Default) for all financialassets (and other elements) not measured at fair value through profit or loss. There are several meth-odologies for estimating this PD from market or historical information. However, in some cases entitiesdo not possess market or historical information concerning a counterparty. For such cases, we propose amodel called Financial Ratios Scoring (FRS), by means of which an entity can obtain a shadow rating fora counterparty as a first step in estimating the PD. The model differentiates from other recent models inseveral aspects, such as the size of the database and the fact that it is focused on non-rated companies, forexample. It is based on scoring the counterparty according to its key financial ratios. The score will placethe counterparty on a percentile within a previously constructed sector distribution using companies witha credit rating published by rating agencies or financial vendors. We have tested the model reliability bycalculating the internal credit rating of several companies (which have an official/quoted credit rating),and by comparing the rating obtained with the official one, and obtained positive results.

©2020 ASEPUC. Published by EDITUM - Universidad de Murcia. This is an open access article under theCC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Códigos JEL:C13G33C63M41

Palabras clave:NIIF 9Deterioro de Activos FinancierosProbabilidad de QuiebraRating Crediticio

Pérdida prevista según la NIIF 9: una propuesta de modelo para la estimaciónde la probabilidad de impago en las empresas sin rating

R E S U M E N

Bajo el modelo de provisiones por riesgo de crédito de la NIIF 9, las empresas deben estimar unaProbabilidad de Default o quiebra (PD) para todos los activos financieros (y otros elementos) no valoradosa valor razonable con cambios en la cuenta de resultados. Existen varias metodologías para estimardicha PD utilizando información histórica o de mercado. No obstante, en algunos casos las empresas nodisponen de información histórica o de mercado acerca de una contraparte. Para estos casos proponemosun modelo denominado Financial Ratios Scoring (FRS), a través del cual la entidad puede obtener unrating interno de la contraparte como primer paso para estimar la PD. El modelo se diferencia de otrosmodelos recientes en varios aspectos como, por ejemplo, el tamaño de la base de datos o el hecho de quese enfoca en empresas sin rating. Se basa en dar una puntuación a la contraparte en función de sus ratiosfinancieros clave. La puntuación sitúa a la empresa en un percentil dentro de una distribución del sectorpreviamente construida utilizando empresas con rating oficial u ofrecido por vendors. Hemos analizado lafiabilidad del modelo calculando el rating interno para empresas con rating oficial y hemos comparado elrating interno con el oficial, obteniendo resultados positivos.

©2020 ASEPUC. Publicado por EDITUM - Universidad de Murcia. Este es un artículo Open Access bajo lalicencia CC BY-NC-ND (http://creativecommons.org/licenses/by-nc-nd/4.0/).

https://www.doi.org/10.6018/rcsar.370951©2020 ASEPUC. Published by EDITUM - Universidad de Murcia. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

https://www.doi.org/10.6018/rcsar.370951

revistas.um.es/rcsar

[email protected]

http://creativecommons.org/licenses/by-nc-nd/4.0/


https://www.doi.org/10.6018/rcsar.370951



D. Delgado-Vaquero, J. Morales-Díaz, C. Zamora-Ramírez / Revista de Contabilidad Spanish Accounting Review 23 (2)(2020) 180-196 181

1. Introduction

Under International Financial Reporting Standards (IFRS)issued by the International Accounting Standards Boards(IASB), accounting rules for financial instruments have re-cently changed. International Accounting Standard (IAS) 39(issued in 1998 and in force since 2001) has been replaced byIFRS 9 for reporting periods beginning on or after 1 January2018 (earlier application was permitted).

IFRS 9 has introduced several changes. For example, thecategories for financial assets are different to those of IAS39 (classification criteria is also different), and changes havebeen made to hedge accounting rules.

One aspect significantly affected by the IFRS 9 changes isloan loss provisioning (impairment rules). For many entities,this has proved to be the most important change (that withthe highest impact). It is not only banks that have been im-pacted by the new impairment rules; in fact all kinds of en-tities are making changes to their provisioning criteria (EY,2018; EY, 2016; Novotny-Farkas, 2016; Beerbaum, 2015;Hronsky, 2010).

The IAS 39 impairment model was based on “incurredlosses”. Several regulators and authorities argued that thismodel led to procyclical effects, and asked standard issuersto develop a new model that entailed a more forward-lookingprovisioning (e.g. BCBS, 2009; FCAG, 2009; FSF, 2009; G20,2009). The new IFRS 9 model is based on “expected losses”instead of “incurred losses”; however, it is not a full expectedloss model.

With certain exceptions1, under IFRS 9 all financial assets2

not measured at fair value through profit or loss should beclassified in three different “stages”. For financial assets in-cluded in stage 1, 1 year expected loss should be estimatedand recognized. For financial assets included in stages 2 and3, expected loss until maturity should be estimated and recog-nized. In other words, for all financial assets (and other ele-ments) subject to IFRS 9 impairment rules, the entity shouldestimate a PD (Probability of Default) for 1 year or matur-ity. The measure of the loan loss allowance will require theuse of data not previously considered under IAS 39 (Holt &McCarroll, 2015, p.20).

IFRS 9 establishes that the estimated PD must include notonly past due information, but also forward-looking inform-ation (in relation to expected changes in default rates). Inthis sense, observed past default rates should be adapted tochanges in macroeconomic variables.

There are several methods for obtaining a PD:

1. If market information of quoted inputs is available, thePD can be directly calibrated from quoted CDS spreads,quoted bonds yields or by using official credit rating andpeer information. In theory, it is assumed that this mar-ket information already incorporates forward-lookingadjustments.

2. A PD can also be obtained by using internal historicaldefault data adjusted by forward-looking information.This data is generally held by large corporate and bank-ing companies.

1For example, purchased or originated credit-impaired financial assets(IFRS 9 paragraphs 5.5.13 and 5.5.14), or trade receivables, contract assetsand lease receivables to which the simplified model is applied (IFRS 9 para-graphs 5.5.15 and 5.5.16).

2IFRS 9 impairment rules do not only apply to financial assets; theyalso apply to lease receivables (under IFRS 16); to contract assets (underIFRS 15); and in many cases to loan commitment and financial guaranteecontracts (IFRS 9 paragraphs 2.1, 4.2.1(c), 4.2.1(d) and 5.5.1).

3. Finally, if no market or internal historical information isavailable, an internal model can be used for estimatingthe PD based on other companies’ default rates, or oninformation from the company’s financial statements orfrom other sources. The models can be split into twogroups:

• Structural models, based on Merton (1974) and onBlack & Scholes’ (1973) option pricing model.

• Non-structural (analytical) models (as Altman etal., 1977).

With regard to the abovementioned third method, severalauthors have proposed internal models for estimating a com-pany’s probability of default. Altman (1968) proposed aninitial analytical model in which he used financial metrics(accounting ratios) for predicting an entity’s default. Otherauthors have proposed structural and analytical models forestimating credit risk or default probability, such as Merton(1974); Kaplan & Urwitz (1979); Ederington (1985); Long-staff & Schwartz (1995); Duffee (1999); and Kamstra et al.(2001).

In this regard, there are also lines of research by otherauthors proposing a model whereby they obtain their owninternal credit rating for a counterparty (also known as an“unofficial” or “shadow” rating). They compare this ratingwith the official credit rating (in order to challenge the offi-cial credit rating). The most recent papers in this area arethose by Creal et al. (2014), Tsay & Zhu (2017), and Jiang(2018). Much part of accounting literature has been dedic-ated to the prediction of business failure (Tascon & Castaño,2012), and it are not been useful for assessing IFRS 9 impair-ments.

Nonetheless, there is a lack of studies focused on non-quoted/non-rated entities. According to Duan et al. (2018),the relative paucity of academic attention is partly due to thelack of publicly available data on privately held firms. Even ifaccounting data for private firms is available, the lack of mar-ket data such as stock prices entails an additional obstacleto studying their defaults, since recent advancements in thecredit risk model typically require some form of market in-formation.

Duan et al. (2018) propose a model for such cases. Theyobtain the distance-to-default (DTD) for quoted companies,and then identify macro and firm-specific factors related tothe DTDs. Subsequently they locate macro and firm-specificvalues for private firms, and utilize the coefficients estimatedfrom public firms to obtain the public-firm equivalent DTDsfor the private firms. In addition, they improve the efficiencyof estimating the default probabilities by adopting the newlydeveloped doubly stochastic Poisson forward intensity modelsuggested in Duan et al. (2012).

Cappon et al. (2018) propose an alternative model whichthey apply to Brazilian banks. They develop a regressionmodel to estimate the “synthetic rating” of Brazilian banksfrom financial variables. They achieve an R2 higher than80% to explain the ratings. However, they do not disclosethe main internal aspects of the model.

Ivanovic et al. (2015) also propose a model for obtaininga “shadow rating”, but it is focused on countries (and not onentities).

Thus, taking the above into consideration, it can be seenthat no credit model has been proposed which includes all ofthe following characteristics:

1. Specifically focused on complying with IFRS 9 expectedloss requirements. The IFRS 9 PD should be based not

182 D. Delgado-Vaquero, J. Morales-Díaz, C. Zamora-Ramírez / Revista de Contabilidad Spanish Accounting Review 23 (2)(2020) 180-196

only on historical information, but should also considerforward-looking information. By way of example, Alt-man’s and Merton’s models do not incorporate forward-looking information (related to market quotes).

2. Able to be applied to non-quoted/non-rated entities.Previous models have mainly been developed for non-quoted companies (Beaver, 1968; Ohlson, 1980; Camp-bell et al., 2008; Chava & Jarrow, 2004)

3. Comparatively easy to implement, so that entities canuse it in order to comply with IFRS 9 impairment re-quirements. For example, Duan et al. (2018) model isbased on 1,759 default events from 1999 to 2014 froma sample of 29,894 Korean firms, whereas a model witha relatively small database is required.

4. Providing an output of a credit rating in the same scaleas the credit rating issued by official rating agencies, sothat the corresponding PD may be obtained from inform-ation derived from comparable companies (companieswith the same rating and belonging to the same sectorand country).

In this paper, we propose a model for obtaining an internalcredit rating as the first step in estimating a PD compliantwith IFRS 9. The model includes calibration with credit rat-ings issued by agencies/financial vendors of companies in thesame or similar sectors, i.e. the model is calibrated with mar-ket data.

We test our model by comparing its output for entities pos-sessing an official credit rating with agency-issued ratings(Moody’s, Fitch, or Standard and Poor’s). Therefore, we ob-tain a unified framework which incorporates a firm’s specificfeatures along with its sectorial and regional factors, andwhich enables market assessments of credit risk to be incor-porated into the book value of financial assets.

This model could also be used by lenders (banks or othernon-banking lenders) who need to determine whether or notto lend3, and the interest rate they should charge to the bor-rower.

In many occasions these lenders do not have market in-formation about the credit quality of the borrower. The creditquality of the borrower is related to the credit spread (overthe risk-free rate) that the lender should charge to the bor-rower in relation to the loan. One possibility is to use themodel that we will propose to obtain an internal rating forthe borrower as a first step for assigning a credit spread.

Once the internal credit rating is obtained the credit spreadcould be estimated, for example, using quoted bond yieldsof peer companies i.e. companies with same rating (officialrating) and same sector / country.

The remainder of the paper is organized as follows: in Sec-tion 2 we introduce IFRS 9 impairment rules and the need fora PD estimation (that may be carried out via a credit rating).In Section 3, we detail the basics of the proposed model, andin Section 4 we describe the model’s methodology. In Section5, we test the model’s results, and in Section 6 we present ageneral conclusion.

2. IFRS 9 impairment rules and PD estimation

As seen in the previous section, IFRS 9 impairment rulesare based on an expected loss model (in contrast with the IAS

3Apart from credit risk, there are other factors that lenders must gen-erally consider when deciding whether or not to give a loan to a certaincounterparty. For example: market or regulatory variables.

39 incurred loss model). All financial assets subject to IFRS9 impairment rules (with certain exceptions), are classifiedin three different stages. Depending on the stage the impair-ment calculation is based on 1 year expected loss (“12-monthexpected credit losses”), or on expected loss until maturity(“lifetime expected credit losses”).

In theory, all financial assets are included in stage 1. Theyprogress to stage 2 when “credit risk on that financial in-strument has increased significantly since initial recognition”(IFRS 9 paragraph 5.5.3). Finally they are classified as stage3 when the loss is incurred.

IFRS 9 impairment rules apply to:

• Financial assets (debt instruments) measured at amort-ized cost or fair value through other comprehensive in-come.

• Lease receivables (IFRS 16).

• A contract asset (IFRS 15).

• A loan commitment and a financial guarantee contractto which the impairment requirements apply in ac-cordance with IFRS 9 paragraphs 2.1(g), 4.2.1(c) or4.2.1(d).

The general formula for impairment estimation is as fol-lows (EY 2019, p.3753):

Ex pected Loss = EADt · PDt · LGDt ·DF(0, t) (1)

Where:

• EADt : represents the Exposure at Default at time t. Thisis the entity’s risk exposure at the time of default. Inother words, the sum that the counterparty owes to theentity at the time of default.

• PDt : represents the Probability of Default at time t.

• LGDt : represents the Loss Given Default at time t. LGDis calculated as 1 less the expected recovery rate.

• DF(0, t): represents the discount factor from the calcu-lation date to t. For calculating the discount factor, theinstrument’s effective rate is used.

t is 1 year in stage 1 (or less than 1 year if the instrumentmatures in less than 1 year), and in stages 2 and 3, it is thetime in years until maturity. t can also be divided into sub-periods (always taking into consideration all the periods untilthe instrument’s maturity in stages 2 and 3). In fact, in thecase of an amortizing loan, it is more correct to divide t intosubperiods, and to use conditional PDs instead of using PDuntil maturity.

LGD value depends on several factors: the value of thecounterparty net assets; the loan’s seniority; and the valueof any specific guarantee. In practice, if no information isavailable, LGD is assumed to be 60% (the recovery rate being40%).4

In the following table we show average corporate debt re-covery rates measured by trading prices from 1983 to 2017(obtained from Moody’s 2018):

Altman et al. (2005) find that recovery rates are a func-tion of supply and demand for securities, with default ratesplaying a pivotal role.

4See Ou et al. (2016) and Koulafetis (2017) for an empirical study ofaverage recovery rates according to collateral.


Table 1Recovery Rates

Source: Moody’s 2018.

Conversely, in order to estimate the PD, we can highlightthe following methods:

1) In theory, the best method of obtaining a counterpartyPD (from a market perspective) is to infer it from a CDS(Credit Default Swap) spread over that counterparty (assum-ing that the CDS is quoted in an active market). This is due tothe fact that the main influence on the CDS spread is the PDof the name behind the CDS. In fact, a company’s credit de-fault swap spread is the cost per annum for protection againstthe severity loss caused by the company default (Hull et al.,2004).

The process for calibrating the PD from CDS spreads is asfollows (Schönbucher, 2003):

Widely used PD models are understood to follow anintensity-based process. N: an event probability with an oc-currence rate λ for a time period T − t =∆t. Namely,

P [N (t +∆t)− N (t) = 1] = λ∆t (2)

so that

P [N (t +∆t)− N (t) = 0] = 1−λ∆t (3)

We subdivide the interval [t, T] into n subintervals oflength ∆t = (T t)/n. In each of these subintervals, the pro-cess N has a jump with probability λ∆t. We conduct n in-dependent binomial experiments each with a probability ofλ∆t for a “jump” outcome. Therefore, the probability of nojump at all in [t, T] is given by

P [N (T ) = N (t)] = (1−λ∆t)n =�

1− 1nλ (T − t)�n

(4)

Since�1− x

n

�n → ex i f n→∞, this converges to a Pois-son process with no event between each subinterval n:

P [N (T ) = N (t)] = e(−λ(T−t)) (5)

Translated into default probabilities, and considering dif-ferent occurrence (hazard) rates λn for different predefinedtime intervals [Tn−m; Tn] of the instrument’s life, the instru-ment’s survival probability between t and t +∆t is:

SP [t, t +∆t] = e(−λi∆t) (6)

And therefore, the cumulative PD in the same context willbe:

PD [t, t +∆t ] = 1− e(−λi∆t) = 1− SP [t, t + ∆t ] (7)

In this framework, the hazard rates for each time buckett, t + ∆t are calibrated by using a standard CDS par pricingmodel:

C redi tSpreadi

n∑i=1

P(0, t i) SPi ∆t =

(1− RR)n∑

i=1

P(0, t i) [SPi − SPi−1] (8)

where P(0, t i) is the risk-free discount factor at time bucketi. The hazard rate λi is calibrated with a given Recovery RateRR depending on the instrument’s seniority in order to com-pute each SP between each time bucket i;i-1.

2) The second method consists of inferring the PD fromquoted bond prices (or bond yields – YTM: Yield To Maturity)(Hull, 2018, p.546 to 526).

The bond’s yield spread is the excess of the promised yieldon the bond over the risk-free rate. The usual assumption isthat the excess yield compensates for the probability of de-fault. However, this assumption is not perfect. For example,the price of a corporate bond is affected by its liquidity: thelower the liquidity, the lower its price.

In practice, the calibration of default probabilities frombond prices can be carried out as follows:

Once the hazard rates have been calibrated following thesteps listed above, in turn a binomial tree may be constructedin order to compute the product value at each tree node, con-sidering the conditional default probabilities [SPi − SPi−1] ateach node i. Namely:

Figure 1Default tree model

Source: Compiled by the authors.

where CF(t i−1, t i) is the risk-free cash-flow to be paid bythe instrument; SP(0, t i) is the survival probability of theproduct between t i−1 and t i; 1−SP(0, t i) is the default prob-ability for the same period; and RR(t i−1, Ti) is the estimatedRecovery Rate for each period.

The tree shows that bond payments do have a survivalprobability at each node t i but it is complemented by its de-fault probability at the same node, where the payment valuewill only be the estimated recovery. That is to say, at everynode t i a default event can occur, or the obligor will continueuntil the next date t i+1. Following default, the non-defaultedpath continues (as indicated by the upper continuation of thetree), but the defaulted security, for a given node, only earnsits recovery payoff and ceases to exist (as represented by thedashed lines). The sum of the payment scenarios (red circles)weighted by the probability of their occurrence is equal to theinstrument’s fair value. The above also means that the prob-abilities attached to the branches of the tree are only the con-ditional default and survival probabilities at this node as seenfrom t = valuation date. Hence under the above model, thedefaultable bond price is:


(9)where CF is the default-free cash flow at each node i.

3) If the above information is not available, other possiblemethods of estimating a PD are using:

• The quoted CDSs spread (over bonds issued by the samecounterparty with the same maturity) in a non-activemarket.

• The quoted YTM of bonds issued by the same counter-party with the same maturity in a non-active market.

• The quoted CDSs spread (over bonds issued by the samecounterparty with similar maturity) in a non-active mar-ket. The spread should be adjusted for the difference inmaturity.

• The quoted YTM of bonds issued by the same counter-party with similar maturity in an active or non-activemarket. The PD is adjusted for the difference in matur-ity.

4) If the specific counterparty does not have quoted CDSsof bonds (in active or non-active markets), the PD may alsobe obtained from the quoted CDSs or bonds of other compan-ies with the same rating and characteristics (sector, country,size, etc.).

5) If the specific counterparty does not have quoted CDSsof bonds, nor a public credit rating, the entity could internallyestimate a credit rating for the specific counterparty. The ob-jective of the model that we present in the following sectionis to estimate that internal credit rating.

Once the credit rating is obtained, there are two possibilit-ies for estimating the PD:

• Derive the PD from the quoted CDSs or bonds of com-panies with the same rating and similar characteristics(sector, country, size, etc.).

• Use the default studies published by the rating agencies(Ou et al., 2016 and Vazza et al., 2018), which showrealized default rates for various rating categories andthe “ratings migration” or transition matrices (see Table2). However, this second possibility does not includeforward-looking adjustments.

Table 2Corporate 1 year default probabilities

Source: Ou et al. (2016) and Vazza et al. (2018).

On the other hand, as stated before, under IFRS 9 the es-timation of future credit losses depends on the classificationof the financial asset in three possible stages. If the financialasset is classified in stage 1, one year expected losses are es-timated (i.e. one year PD is used). If the financial asset isclassified in stages 2 or 3, lifetime expected losses are estim-ated (i.e. PD until maturity is used).

Initially, all financial asset are classified in stage 1. A fin-ancial asset is reclassified to stage 2 if there has been a sig-nificant increase in credit risk since initial recognition. If thefinancial asset is impaired (credit losses are incurred), thefinancial asset is classified in stage 3.

At each reporting date, an entity shall assess whether thecredit risk on a financial instrument has increased signific-antly since initial recognition. When making the assessment,an entity shall use the change in the risk of a default occur-ring over the expected life of the financial instrument insteadof the change in the amount of expected credit losses. Tomake that assessment, an entity shall compare the risk of adefault occurring on the financial instrument as at the report-ing date with the risk of a default occurring on the finan-cial instrument as at the date of initial recognition and con-sider reasonable and supportable information, that is avail-able without undue cost or effort, that is indicative of signi-ficant increases in credit risk since initial recognition (IFRS 9paragraph 5.5.9).

IFRS 9 (paragraph B5.5.17) recognizes that a significantchange in the instrument’s credit rating is a possible way toanalyze if there has been a significant increase in credit risksince initial recognition. Therefore, FRS model can also beused for this.

As an exception to the stages, an entity shall always meas-ure the loss allowance at an amount equal to lifetime expec-ted credit losses for:

(a) trade receivables or contract assets that result fromtransactions that are within the scope of IFRS 15, andthat:

(i) do not contain a significant financing componentin accordance with IFRS 15 (or when the entityapplies the practical expedient in accordance withparagraph 63 of IFRS 15); or

(ii) contain a significant financing component in ac-cordance with IFRS 15, if the entity chooses as itsaccounting policy to measure the loss allowance atan amount equal to lifetime expected credit losses.That accounting policy shall be applied to all suchtrade receivables or contract assets but may be ap-plied separately to trade receivables and contractassets.

(b) lease receivables that result from transactions that arewithin the scope of IAS 17, if the entity chooses as itsaccounting policy to measure the loss allowance at anamount equal to lifetime expected credit losses. Thataccounting policy shall be applied to all lease receivablesbut may be applied separately to finance and operatinglease receivables.

3. Proposed credit model (model basics)

Our model may be called the “Financial Ratios Scoring”(FRS) model. We have developed it ourselves, and we haveused it in practice during consultancy services for a widerange of companies and sectors.


The FRS model is partially based on Duan et al. (2018)and Ivanovic et al. (2015).

Its main model input is the information obtained throughthe counterparty financial statements (i.e. the main inputsare financial ratios). According to the values of several keybalance sheet and profit and loss account ratios, the companyis allocated to a certain position (score) within a consistentdistribution of companies that possess an official credit rating(issued by a rating agency or quoted by relevant financialvendors), and which belong to the same or similar sectors.The position within the distribution is related to a certaincredit rating (the official rating of companies with a similarscore).

According to Cappon et al. (2018), credit ratings are “opin-ions” issued by rating agencies regarding the credit worthi-ness of corporate, municipal and sovereign borrowers. Agen-cies generally avoid claiming that credit ratings predict prob-abilities of default. Nevertheless, they do publish detaileddefault studies which show historical ratings migration anddefault events as a function of the initial rating and time hori-zon. Analysts and risk managers routinely use default studydata as estimates of default probabilities. In practice, it isassumed that a rating generally matches a range of defaultprobabilities.

The FRS model is intensive in terms of data collection (aswe will see, it is necessary to create a distribution of sectorcompanies). However:

• It can be considered to be highly consistent since themodel’s inputs are calibrated with the financial inform-ation of companies which do have an agency rating.

• It is not as intensive as other similar models (see section1).

Among the ratios considered by the model (which may alsovary from one sector to another), those with higher relevancein terms of credit risk are those related to debt and interestcoverage, leverage or liquidity. In other words, the relativedebt level of a company is generally the factor with most in-fluence on its credit risk. Growth and profitability are alsoconsidered, but linked to liabilities and equity.

As the FRS relies on accounting information, two possiblelimitations to the model are earnings manipulation, and thefact that qualitative information is not considered.

With regard to earnings manipulation, Alissa et al. (2013)identify firms that deviate from expected credit ratings, anddemonstrate that these empirically estimated credit rating de-viations are associated with earnings management activities.Their results suggest that firms above or below their expectedcredit ratings may be able to successfully achieve a desireddowngrade or upgrade through the use of earnings manage-ment. Therefore, if the financial information has been ma-nipulated, the credit rating obtained will also differ from thecorrect rating. Nevertheless, our model generally assumesthat the financial information used is correct. The model’smain objective is not to detect possible fraudulent activity re-lated to financial statements.

Conversely, default risk (credit risk) can generally be meas-ured in three different ways: using quantitative data; usingqualitative data; or by using a combination of both. Quant-itative data includes equity prices; credit market data; fin-ancial instrument quotes other financial data, etc. Qualitat-ive data includes the entity’s structure; how the entity is per-ceived by the market; business estimations; business plans;information regarding the entity’s governance and risk ap-petite, etc.

The model we propose uses quantitative data as its maininput (financial information (financial ratios) obtained fromthe entity’s financial statements). In theory, we do not usequalitative data in our model, fundamentally due to the fol-lowing factors:

• Qualitative factors or metrics are difficult to measureand to model due to several reasons such as the fact thatthe same information is not available for all entities, andthey entail a significant level of subjectivity, etc. In thissense, as our model aims to be both robust and easy toimplement at the same time, it does not consider qualit-ative factors (at least not directly).

• In recent years, financial and market information hastended to be more reliable (quantitative factors). Thismakes quantitative factors more effective when estimat-ing a probability of default or assigning a credit rating.In fact, in terms of default events and recovery rates,quantitative models have been taking new assumptionsinto account and covering recent scenarios (for example,see Moody’s latest reports on default risk and recoveryrates (Moody’s, 2017).

It can be argued that FRS model uses forward-looking in-formation (as required by IFRS 9). This is because, as we willfurther see in the following section, the obtained credit ratingis calibrated using official credit ratings that are developed byofficial rating agencies. These official ratings are obtained bythe agencies not only using quantitative information (finan-cial statements ratios) but also using qualitative factors like:business perspectives, corporate governance, the regulatoryand competitive environment, financial policy, etc.

Therefore, qualitative aspects are also covered to a certaindegree. However, those aspects are not directly modelled normeasured.

Moreover, once the credit rating is obtained for a certaincounterparty, the second step (obtaining the PD) can be done(as stated before) by reference to quoted CDSs of bonds. Thisinformation includes, in theory, all available market forward-looking information

If the PD is obtained using historical default rates, forwardlooking information should me incorporated. For example,the PD can be adapted to current macroeconomic conditionsvs. conditions that existed when the data was collected.

4. Methodology, risk factor calibration and implement-ation

As previously stated, the FRS model is based on reflect-ing the position (score) of a company within a representativegroup of rated companies, so as to provide the company witha credit rating in line with its associated score.

With regard to this score:

• It may also be considered as a “percentile” in the modelcontext (in fact, it is a percentile within a sectorialgroup). It is configured on a basis where 1 representsthe worst position and 100 the best position.

• It will depend on the values of the financial ratios selec-ted, and therefore on the position of each financial ratiowithin its group (hereinafter “distribution”).

The construction of the model consists of five main phases:

• Phase 1 – Definition of key financial ratios


• Phase 2 – Database and general score

• Phase 3 – Specific score for each ratio

• Phase 4 – Scoring matrix and β j calibration

• Phase 5 – Obtaining the credit rating for the company

Phase 1 – Definition of key financial ratios

In this first phase, a group of key financial ratios is definedfor the sector to which the company belongs. Generally, theseratios are related to metrics such as coverage, leverage, li-quidity, profitability and growth.

We propose the use of the ratios shown in Table 3 belowas a general framework. These ratios are widely used byanalysts (Fazzini, 2018) and by rating agencies (see Moody’s2017b, for example), since they represent the key financialdimensions that act as drivers for a rating profile. They areeasy to calculate using the financial information included inthe public financial statements issued by companies.

Nevertheless, it should be noted that additional ratioscould be included for specific sectors according to the natureof their business, such as Passenger Load in the commercialairlines sector, or Loan Default Rate in the banking sector,etc. As will be explained in subsequent sections, the intrinsiccharacteristics of a sector are disclosed when calibrating theratio weights, hence to a certain extent the “sector” variableis covered by this methodology.

Table 3Ratios used in the FRS model


• “Interest expense/Sales” and “EBITDA/Interest Ex-pense” (coverage ratios) focus on to what extent interestexpenses related to debt are “covered” by income fromnormal business operations. The higher the interest ex-pense in relation to sales or EBITDA, the weaker the fin-ancial position of the company. In other words, the ratioanalyzes to what extent the entity generates sufficient re-sources in order to be able to pay the interests relatedto external debt.

– In the first ratio, the higher the level, the lower thecoverage (less sales income is available to pay theinterest expense).

– In the second ratio, the higher the ratio level, thehigher the generated surplus (and the higher thecoverage).

In general terms, the model places much import-ance on coverage ratios as a default event is usuallyunderstood as the situation in which a company isnot able to entirely pay the short-term debt. In this

sense, coverage ratios can act as signals of creditproblems.

• “(Liabilities - Cash & Securities)/Assets” statically ana-lyzes the company’s leverage level (or relative debtlevel). It compares the assets (that could be used to paythe debt) with the net debt (net of cash and liquid secur-ities). The higher the ratio result, the higher the relativedebt level (and the higher the credit risk).

• “Retained Earnings/Liabilities” compares the company’sresult with its debt level. It analyzes the company’s lever-age level more dynamically. The higher the ratio result,the lower the credit risk.

• “Current Assets/Current Liabilities” is known as work-ing capital. Depending on the sector involved, the inter-pretation of the result may vary. Generally speaking, thehigher the ratio, then the higher the liquidity level. Nev-ertheless, in the retail sector, a low ratio may be inter-preted in a positive way, i.e. the entity is being financedby its suppliers (the average collection period is lowerthan the average payment period).

• “Cash & Securities/Current Assets” analyzes to what ex-tent current assets are composed of liquidity (the higherthe ratio level, the higher the liquidity level).

• ROA and ROE analyzes the profitability of the company.They calculate return in relation to assets (ROA), andreturn in relation to equity (ROE).

• “Sales growth” analyzes the growth in sales figures.Company growth can be analyzed in several ways (interms of assets, sales, EBITDA or Net Profit among oth-ers). We have chosen sales growth given the fact thatthe other metrics may be biased due to the company’sactivity. Sales figures are usually sufficiently isolated tobe considered as a good estimate of company perform-ance (always considering their relevancy in comparisonto the abovementioned credit metrics).

Phase 2 – Database and general score

This phase consists of creating a database including a port-folio of companies (“peers”) which possess an official creditrating (issued by a rating agency), and of giving a generalscore to each of them.

Where possible, companies included in the databaseshould belong to the same sector and country as the com-pany under analysis, and should have recently been rated bya relevant credit rating agency (i.e. Moody’s, Fitch or S&P).Alternatively, given the limited number of rated companiesover the sectors, the database can also be created by usingthe credit rating issued by Reuters or Bloomberg. The use ofthe Reuters or Bloomberg credit rating has the added advant-age of giving a score between 0 and 100 as well as the rating.A given score level is related to a rating level.

In some cases, it can prove difficult to find peers since com-panies are highly diversified and act in many different indus-tries and markets at the same time. Nevertheless, we recom-mend the inclusion of as many peers as possible.

A score is assigned to each peer company in the portfo-lio, and each company is ranked according to its position (apercentile between 1 and 100) within the entire portfolio ofcompanies. This position represents the general score. Byway of example, for a specific real case (in a specific sector),


Figure 2Example of Score distribution per Credit Rating


we included the information required in a database in orderto build the cumulative distribution function (Figure 2).

As can be seen in the above distribution (Figure 2), thecredit rating is directly related to the score (“position” or “per-centile”) within the distribution. This distribution was cre-ated using Reuters. Certain companies with an equal ratingare scored slightly differently according to their outlook, sizeand debt coverage. In the example above, this means thatthere are 13 companies with the same rating (BBB-) betweenpercentile 25 and 37. This is normal given the fact that thereare more companies rated between BBB- and BBB+ than inany other rating bucket. Figure 2 above represents a cumu-lative distribution of 63 rated companies. Its correspondingprobability density function is shown in Figure 3:

Figure 3Example of Density function of Credit Rating


Phase 3 – Specific score for each ratio

In this phase, we assign a score to each peer company inthe portfolio in relation to each ratio. In other words, eachcompany has on the one hand a general score (Phase 2), andon the other a specific score for each ratio (Phase 3).

Therefore:

• We calculate every ratio included in Table 3 for all of thecompanies in the sample.

• We create a distribution for each ratio according to theresults.

• Each company is given a score (percentile) for each ratiodepending on its position within the distribution.

In theory, the ratio distribution (vector) should be as gran-ular as possible.

Phase 4 – Scoring matrix and β j calibration

A matrix is prepared which shows the relationship betweenthe comparable companies’ rating, their general score, andthe score of each ratio. The following table presents an ex-ample:

Table 4Example of Scoring table (general and specific scores)


The above matrix (Table 4) should be fed with the follow-ing inputs:

• The name of each peer company in the portfolio.

• The rating and general score of each peer company inthe portfolio.

• The specific score of each peer company in the portfoliofor each ratio.

We can also obtain a score for each ratio for the companyunder analysis (using the previous distribution). The ques-tion is how we can use the scores for each ratio for the com-pany under analysis in order to assign a general score for thatcompany (and, therefore, a rating).

The first item to be calculated is the representativeness ofeach ratio within the rating assigned to each company, i.e. towhat extent the score of each ratio influences the generalscore. We know that the ratios used do not entirely cover thewide range of risk factors considered by rating agencies (al-though they do implicitly include qualitative factors). Giventhe nature of our analysis, it is clear that the overall creditscore is a dependent variable, and that the ratios’ scoresare the independent variables, assuming there are risks notcovered in the model. From our previous practical research(see also the empirical text in the following section), it canbe concluded that the use of an Ordinary Least Squares meth-odology to calibrate a linear regression represented by aweighted sum of the ratios scores gives highly accurate res-ults in terms of the model’s goodness of fit. In other words,we are able to estimate the overall credit score of a companyas follows:

Scorecompanyi=

n∑j=1

Scorecompany jβ j (10)

Where β j is the coefficient assigned to the ratioj. β j is cal-culated using the database of companies already created. Byway of example, we use the data included in Table 5 (an ex-ample of a Scoring Matrix) in order to calibrate each β j for agiven sector:

Following a Least Squares methodology, the following β j5

are obtained with regards to (10):

5For model calibration, β jcalibration via the OLS method can, for in-stance, be bounded between 0.01 and 0.99, with a total sum of 1 with re-


Table 5Scoring table to be used as an example


Table 6β j calibration for (3) with data set of Table 5

Table 6 𝛽𝑗 calibration for (3) with data set of Table 5

Profitability Leverage Coverage Liquidity Growth

β1=5.45% β2=42.27% β3=48.03% β4=3.25% β5=1.00%

Source: Compiled by the authors. Source: Compiled by the authors.

The linear regression and the R2 for the previous calibra-tion are shown below. The representativeness and goodnessof fit of the model are sufficiently satisfactory to be able toconsider the model as consistent, as no high multicollinearityis found. It should be noted that not all sectors fit equally ina linear model, and the dependency on the database size andon certain ratios is relatively moderate. The analyst shouldselect the sectorial ratios which best represent the credit per-formance.

Figure 4Example of Linear regression plot for the FSR model example


The representativeness and goodness of fit of the model

sidual explanatory strength loss for simplification purposes. Usually, thetotal sum of calibrated betas via OLS for this model is somewhat higherthan 100%, which may lead to final percentiles higher than 100. Also, thefloor in 0 is used to avoid multicollinearity issues between inversely correl-ated variables when many explanatory variables (ratios) are used. However,it is highly convenient to firstly calibrate the model with no boundaries inorder to analyze the statistical strength of the model, so that extremely non-significant variables and those entailing multicollinearity issues can be elim-inated, and homoscedasticity and normality in residuals can then be assured.This subject is covered in Section 5.

appear to be consistent with general market ratings.

Phase 5 – Obtaining the credit rating for the company

Once the β j are obtained, two different methodologies canbe applied so as to assign a general score (and therefore acredit rating) to the company under analysis.

The first methodology consists of applying the model as (3)in order to obtain the general score of the company. If, forexample, we consider that the company (the counterparty)analyzed has the following ratio scores:

Table 7Ratio Scores of the analyzed company

Table 7 Ratio Scores of the analyzed company

Profitability Leverage Coverage Liquidity Growth

24 19 38 32 56

Source: Compiled by the authors. Source: Compiled by the authors.

then the model gives a score of 29.19, which means a BBB-rating in line with the score distribution shown in Figure 1.

The second methodology consists of applying a solutionto the model based on a difference-simulation methodology,which in turn covers the root mean square error as far asis possible, which in this example was 6.25. That is to say,while taking into account the existing convexity in the rela-tionship between a company’s general score and its impliedcredit rating, we propose that the weighted sum of differ-ences between the analyzed company’s ratio scores and thoseof each comparable company should be computed. Firstlywe obtain the distance between the analyzed company andthe comparable company. Thus the simulated company scorewill be equal to the sum of the weighted sum of differencesand the current comparable company score.

Scorecompany|comparable i = n∑j=1

(ScoreCompany ratio j− ScoreComparablei ratio j

) β j

+Scorecomparablei

(11)

This is carried out in order to capture the actual differencebetween our computed rating and the theoretical rating thatthe analyzed company would have if a starting point weretaken. That is, we compute the weighted difference basedon the calibrated β j but, by applying the difference to theactual score of the comparable company, we place the ana-lyzed company in a score according to a central point. In thisway, the regression error is covered to a certain degree. Eachresult may be considered as a simulation. The average of sim-ulations will represent the company’s score. The simulationplot can be seen below, and retrieves a concentration abovea score of 32 (BBB-), with an average of 31.76 and a medianof 32.84. This method also provides us with an estimate asto the range in which the score can be placed. It would ob-viously be necessary to include many more companies in thedatabase in order to perform a consistent simulation, but forthe sake of clarity, this example has been carried out from thesample listed in Table 5.

5. Empirical test

So as to assess the accuracy of the model proposed, wehave selected several companies within a given sector in or-der to apply the model to them. The outcome is expected


Figure 5Company score simulations performance


to be equal to the overall ratings, or much closer to thoseprovided by the market.

The chosen sector is “Global Surface Transportation andLogistics”. This sector needs some further ratios to be addedto those detailed in Section 3. In fact, certain ratios needto be eliminated in order to avoid multicollinearity issues orbetas near to zero.

The ratios added are detailed below. They improve thefinancial representativeness within the model since they areexpected to explain particular characteristics of the sector(Moody’s (2017c) and Moody’s (2017d)):

• Profitability, measured by pre-tax income as a percent-age of sales. It provides a metric on the company’s ef-fectiveness with regard to the cost structure and its cap-ability to reach yield premiums in comparison with peercompanies. The capital-intensive nature of the trans-portation industry makes it important to include interestexpense when considering profitability, as capital costsare as relevant as operating costs. Therefore, while thisratio may be relevant for modelling purposes, correla-tion and significance should also be checked.

• Financial leverage and coverage metrics are indicatorsof a company’s financial capacity and long-term viab-ility. Financial flexibility is critical to this sector as itindicates the degree of stress a company would sufferduring an economic downturn. In addition, leverage af-fects a company’s ability to reinvest in the business, as ahighly leveraged company may not have the same accessto capital (new funds) as other companies with a lowerleverage level. Furthermore, leverage partly affects thecapacity to deal with changing market conditions in thehighly cyclical business operations to which this type ofcompany may be exposed. Financial leverage and cov-erage are additionally represented in the model by thefollowing ratios:

– Debt/EBITDA ratio is an indicator of debt ser-viceability and leverage, and is commonly usedin this sector as a proxy for comparative financialstrength.

– Funds from Operations (Free Cash Flows fromOperations minus Capex) to Debt is an indic-ator of a company’s ability to repay principal on itsoutstanding debt. This ratio compares cash flowgeneration from operations before working capitalmovements to outstanding debt.

– EBIT to Interest Expense is an indicator of a com-pany’s ability to cover its ongoing costs of borrow-ing.

Other ratios specifically related to airlines (e.g. PassengerLoad) were also candidates for inclusion, but were finally dis-carded since the database created is heterogeneous in termsof the transportation sector.

In this way, we constructed a wide data set of companiesbelonging to the “Global Surface Transportation and Logist-ics” sector for which overall ratings are issued by Reuters. Wehave computed the following ratios and the inherent percent-ile within the sample for each company and ratio: Pre-taxIncome to Sales; Debt to EBITDA; Funds from Operations toDebt; EBIT to Interest Expense; Return On Equity; Net Mar-gin; Return On Assets; EBITDA to Interest Expense; Debt toEquity; Debt to Assets, Cash to Total Debt; Short Term Debtto Total Debt; and Quick Ratio.

Table 8Sector companies used to test the FRS model

Source: Reuters.

The following companies were finally chosen from thesample in order to calibrate the model factors, since they pos-


Table 9Sector companies and their ratio percentiles from the sample

Source: Compiled by the authors; Reuters.

sess the most liquid, updated financial information in linewith the financial statements date used (31/12/2015). Allof these companies do not exactly move in the same busi-ness lines or even in the main sector. However, one shouldconsider the fact that only including companies fully match-ing in terms of size, sector and business lines can make themodel calibration poor as there exist very few companies ex-actly equal, concerning the above requirements. Therefore, ithas been necessary to widen the sample with similar compan-ies although they do not fully match each other regarding alltheir business lines or their inherent nature. What we havebeen looked for is the construction of a consistent samplewith as many similar companies as possible taking into con-sideration the number of companies with ratio distribution in-formation on financial data bases and vendors (e.g. Reuters).In terms of explanatory capacity, most of these companies aresimilar to the initial sample ones: (Table 8)

The following figure (Table 9) shows the general and par-ticular percentiles for each sample company and ratio withinthe sectorial database used:

It should be highlighted that the overall percentile for eachcompany was assigned by a random number within a confid-ence interval of percentile for its rating, in order to test thecapacity of the model to cover the existing percentile disper-sion within the same notch (which may at times result inrating down/upgrades), particularly for ratings from BB+ toBBB+, as explained below.

Testing the sample distribution:

The randomness upon which a company overall percentilemay vary was applied by using the following sample distribu-tion:

Figure 6In - sample distribution function

Source: Compiled by the authors; Reuters.

It is assumed that the percentile distribution follows a nor-mal distribution function, therefore the average expected dis-persion within each rating notch has been applied followinga normal distribution. The figure below shows a sample ofthe simulation6 for the initial percentile of each company:(Figure 7)

It may be noted that generally speaking, higher dispersionoccurs in notches near to non-investment grade letters orbelow, with both the highest investment grades and highlyspeculative grades/defaulted remaining almost unchanged,

6The simulation has been performed assuming that the percentile dis-tribution follows a normal density function. This assumption is taken fromthe empirical distributions, e.g. shown in figures 3 and 6. Therefore, thestandard deviation is computed for each single notch, and assuming alsothat each notch percentile is also normally distributed, this percentile is sim-ulated with a given standard deviation (in percentiles from the average) andaverage, as per the empirical distribution.


Figure 7Percentile dispersion simulation from Probability Density Function


as expected according to the most recent average migrationmatrices:

Table 102017 One-Year Letter Rating Migration Rates

Source: Moody’s; compiled by the authors.

Source: Moody’s compiled by the authors.

Certain specific percentile simulations were tested with thefollowing results:

Table 11Simulated vs historical notch migration probability

Source: Reuters; Moody’s; compiled by the authors.

Thus we can assume that the distribution used, built uponthe overall percentile distribution as taken from Reuters, isreliable for modelling purposes.

Once the overall percentile distribution taken from the sec-torial sample is found to be correct, the percentiles for allthe ratios are computed, and the betas are calibrated un-der OLS following the abovementioned methodology. Firstly,as mentioned in footnote 4, the model is calibrated withoutboundaries so as to check statistical strength and eliminatenon-significant variables. Thus many of the ratios previouslydefined in Phase 1 are eliminated given the multicollinearityin the regression, since there are high correlations (measuredin differences to avoid stationarity issues) between them andthe new explanatory ratios. No relevant information from

certain other ratios is added to the model for this particularsector (betas in the regression are near to 0 or negative dueto the inverse correlation and non-significance).

Thus in this case, Funds from Ops to Debt and EBIT to In-terest Expense are found to be the ratios with relevant statist-ical significance (95% confidence, 13 degrees of freedom) asfollows

Table 12Significance statistics for the main ratios


while Pretax Income/Sales, Debt EBITDA Debt/Assets arethe variables with residual explanatory strength. The correl-ation matrix presented below helps to explain the multicol-linearity issues to be avoided:

Meanwhile, the regression betas with no boundaries withthe most explanatory variables are as follows:

Table 13Ratios and their corresponding calibrated betas for the tested sector


The homoscedasticity, multicollinearity and normality inresiduals have been also analyzed with the following results:

• Multicollinearity:

Table 14R2 and VIF for the two significant explanatory variables



Figure 8Ratio percentiles’ correlation matrix (measured in differences)


• Homoscedasticity: Breusch-Pagan test outcome with ap-value = 0.897.

• Normality in residuals: Jarque-Bera test with a p-value> 0.70 with 100,000 simulations. Although the sampleis relatively small, we also get the following QQ plot:

Figure 9QQ Plot for normality test in regression residuals


Hence the calibration provides an R2 of 83.22%between the actual percentile for each companyand the one simulated:

The previous calibration provided a sum of betas equal to104.75%. If the boundaries for making the sum of betas equalto 100% are applied, the results are almost unchanged:

Table 15Ratios and their corresponding calibrated betas for the tested sector, withboundaries


R2 in this case remains almost the same: 83.07%.

Figure 10Regression plot (Modelled vs Actual percentiles)


In order to check the goodness of fit added to the modelwhen including the new explanatory variables, we also calib-rated the model with the standard ratios as defined in Phase1. The results are shown below:

Table 16Standard ratios and their corresponding calibrated betas for the testedsector, with boundaries


with Return on Equity, Cash to Total Debt and Short-TermDebt to Total Debt with null explanatory strength in any ofthe local minimums found when calibrating the model.

As can be seen, once specific-sectoral ratios are used, theexplanatory capacity added to the model is boosted by at


Figure 11Regression plot (Model vs Actual percentiles), Standard ratios


least 11% (increasing from an R2 of 72% to 83%).It should be noted that including the whole set of ratios

(general + sector-specific) in the regression model only in-creases the explanatory power from 83% to 87.92%. How-ever, the multicollinearity added to the model is huge. Hence,the idea of using sector-specific ratios to build the model isassumed as the basis for this kind of models.

The model calibration captures the amount and directionof trend changes for each company percentile in generalterms, with a correlation measured in levels and in differ-ences higher than 90%. Hence the financial hypotheses car-ried out for the sector outperform regular ratios:

Official Rating vs model rating

The following table compares the model’s rating for com-panies with agency and vendor ratings:

Table 17Comparison between FRS model outputs, agency ratings and Reutersrating (31/12/15)

Source: Moody’s; S&P; Reuters; compiled by the authors.

Out-of-sample testing:

In order to test the predictive ability of the model, we ran-domly selected 4 companies belonging to the tested sectorpossessing an overall percentile published in Reuters, and weapplied the model to them once the financial ratios and theirpercentiles were computed:

The results obtained when using the sectorial calibrationare shown below:

As can be seen, the model’s ability to predict the expectedrating, in this case as quoted by Reuters, can be considered tobe reliable. The sensitivity of the outcomes to the percentiledistribution and its link to several rating letters (e.g. Norwe-gian Air Shuttle) should be noted. However, in the majority

of cases, the difference between the actual rating and themodel rating is null or only one notch. Therefore, it does al-low a link to be made between a given rating and the PDsquoted by rating agencies or financial vendors for the pur-pose of IFRS 9 impairment computation.

6. Conclusion

Under the IFRS 9 impairment model, entities must estim-ate the probability of default for all counterparties from fixedrate financial assets (loans, receivables, bonds, etc.) andother elements not measured at fair value through profit orloss. There are several means of obtaining the correspondingPD and generally speaking, banks and other large companieshave developed internal methodologies.

In other cases, however, entities face difficulties in assign-ing a PD to certain counterparties due to the lack of marketinformation concerning the counterparty; to the lack of in-ternal or external historical default information; and to thelack of a scoring model.

In this paper, we propose a model for such cases (FinancialRatios Scoring – FRS) for which we have detailed 5 steps orphases. Fundamentally it consists of assigning a score to thecounterparty according to its key financial ratios. The scoreplaces the counterparty on a percentile within a previouslyconstructed sector distribution using companies with an offi-cial credit rating.

As we have seen, the model has several advantages:

• The only information needed for the counterparty isfinancial information (financial ratios) that can be ob-tained from financial statements.

• The database to be constructed is relatively small as com-pared to databases required by equivalent models.

• The model output is calibrated with market information(official credit ratings issued by rating agencies or fin-ancial vendors). Therefore, it somehow incorporatesforward-looking information. Moreover, if quoted CDsof bonds are used for obtaining the PD (in a second step)updated market information is used.

• The model is relatively easy to implement.

Regarding the first of the abovementioned advantages, itshould be noted that the information obtained from the coun-terparty’s financial statements may not be of use if it is veryold, or if an event has occurred (general or specific) since thedate of the financial statements which may have changed thecounterparty’s creditworthiness.

The results of the empirical test performed show that themodel works when certain information is available, and thatthe β j coefficients are able to explain the general score witha high degree of confidence.

Moreover, it should be outlined the necessity for modelmonitoring and control. Due to the fact that reported inform-ation and therefore, percentile distribution could changeover time. Reporting events affect the financial informationso that ratios and figures change, so that the model shouldbe reviewed and recalibrated frequently. Also, databasesshould be monitored, as companies’ business and structuremay change as they try to adapt to business and economicconditions, including mergers, divestments, economic down-turns, etc. Subsequently, the model user should monitorthe explanatory variables in terms of representativeness andeven include new ones if the business structure requires newmetrics.


Figure 12Actual percentiles vs simulated percentiles for the sample chosen


Table 18Tested companies and their ratio percentiles from the sample


Figure 13Actual percentiles & ratings vs Simulated percentiles & ratings for the

tested companies


Funding

This research did not receive any specific grant from fund-ing agencies in the public, commercial or not-for-profit sec-tors.

Conflict of interests

The authors declare no conflict of interests.

7. References

Alissa, W., Bonsall, S.B., Koharki, K., & Penn, M.W. (2013).Firms’ use of accounting discretion to influence their creditratings. Journal of Accounting and Economics, 55, 129–147.https://doi.org/10.1016/j.jacceco.2013.01.001

Altman, E.I. (1968). Financial Ratios, Discriminant Ana-lysis and the Prediction of Corporate Bankruptcy. Journalof Finance, 23 (4), 589-609. https://doi.org/10.1111/j.1540-6261.1968.tb00843.x

Altman, E.I., Haldeman, R.G., & Narayanan, P. (1977).ZETATM analysis. A new model to identify bankruptcy riskof corporations. Journal of Banking & Finance 1 (1), 29-54

Altman, E. I., Brady, B., Resti, A., & A. Sironi (2005). ’TheLink Between Default and Recovery Rates: Theory, Empir-ical Evidence, and Implications. Journal of Business, 78, (6),2203–2228. http://dx.doi.org/10.1086/497044

Basel Committee of Banking Supervision (BCBS). (2009).

https://doi.org/10.1016/j.jacceco.2013.01.001

https://doi.org/10.1111/j.1540-6261.1968.tb00843.x


http://dx.doi.org/10.1086/497044


Guiding principles for the replacement of IAS 39. Availableat: http://www.bis.org/publ/bcbs161.pdf.

Beaver, W. H. (1968). Alternative accounting measures aspredictors of failure. The accounting review, 43(1), 113-122.https://www.jstor.org/stable/244122

Beerbaum, D. (2015). Significant Increase in Credit RiskAccording to IFRS 9: Implications for Financial Institu-tions. International Journal of Economics & Management Sci-ences. Available at: https://pdfs.semanticscholar.org/b700/d412674ceae6030745437abe057894dc1cd6.pdf

Black, F., & Scholes, M. (1973). The Pricing of Optionsand Corporate Liabilities. Journal of Political Economy, 81(3), 637-654. http://www.jstor.org/stable/1831029

Campbell, J., Hilscher, J., & Szilagyi, J. (2008). In searchof distress risk. J. Finance 63 (6), 2899–2939. http://dx.doi.org/10.2139/ssrn.770805

Cappon, A., Gorenstein, A., Mignot, S., & Manuel, G.(2018). Credit Ratings, Default Probabilities, and Logar-ithms. Journal of Structured Finance, 24 (1) 39-49. https://doi.org/10.3905/jsf.2018.24.1.039

Chava, S., & Jarrow, R., (2004). Bankruptcy predictionwith industry effects. Rev. Finance 8 (4), 537–569. https://doi.org/10.1093/rof/8.4.537

Creal, D. D., Gramacy, R. B., & Tsay, R. S. (2014).“Market-based Credit Ratings”. Journal of Business and Eco-nomic Statistics, 32 (3), 430-444. https://doi.org/10.1080/07350015.2014.902763

Duan, J.C., Sun, J., & Wang, T. (2012). Multiperiod cor-porate default prediction – A forward intensity approach.Journal of Econometrics 170, 191–209. https://doi.org/10.1016/j.jeconom.2012.05.002

Duan, J.C., Kim, B., Kim, W., & Shin, D.(2018). “Default Probabilities of Privately HeldFirms”. Journal of Banking and Finance, 94, 235-250.https://doi.org/10.1016/j.jbankfin.2018.08.006

Duffee, G.R. (1999). “Estimating the Price of Default Risk”.The Review of Financial Studies, 12 (1), 197-226. https://doi.org/10.1093/rfs/12.1.197

Ederington, L.H. (1985). Classification Models and BondRatings. Financial Review, 20 (4), 237-262. https://doi.org/10.1111/j.1540-6288.1985.tb00306.x

EY (2016). Applying IFRS. IFRS 9 for non-financial entities. Available at: https://www.ey.com/Publication/vwLUAssets/Applying/_IFRS/_%E2%80%93/_IFRS/_9/_for/_non-financial/_entities//$FILE/Applying-FI-Mar2016.pdf

EY (2018). IFRS 9 Expected Credit Loss Making sense ofthe transition impact. Available at: https://www.ey.com/Publication/vwLUAssets/ey-ifrs-9-expected-credit-loss//$File/ey-ifrs-9-expected-credit-loss.pdf

EY (2019). International GAAP 2019. Wiley. UK.Fazzini, M. (2018). Business Valuation. Theory and

Practice. Palgrave Macmillan. https://doi.org/10.1007/978-3-319-89494-2

Financial Crisis Advisory Group (FCAG) (2009). Re-port of the financial crisis advisory group. Availableat: http://www.ifrs.org/News/Press-Releases/Documents/FCAGReportJuly2009.pdf

Financial Stability Forum (FSF) (2009). Reportof the financial stability forum on addressing pro-cyclicality in the financial system. Available at:http://www.financialstabilityboard.org/wp-content/uploads/r/_0904a.pdf

G20 (2009). London summit – Leader’s statement 2 April2009. Available at: https://www.imf.org/external/np/sec/

pr/2009/pdf/g20/_040209.pdfHolt, O., & McCarroll, J. (2015). IFRS 9 not just for banks,

you know! Accountancy Ireland, 47 (3), 18-20.Hronsky, J. (2010). IFRS 9 impairment and procyclicality:

is the cure worse than the disease?. JASSA The Finsia Journalof Applied Finance, 4, 55-59.

Hull, J.C. (2018). Options, Futures and Other Derivatives.Tenth Edition. Pearson. New York.

Hull, J. C., Predescu, M., & White, A. (2004). Relation-ship between Credit Default Swap Spreads, Bond Yields, andCredit Rating Announcements. Journal of Banking and Fin-ance, 28, 2789–2811. https://doi.org/10.1016/j.jbankfin.2004.06.010

Ivanovic, Z., Bogdan, S., & Baresa, S. (2015). Modelingand Estimating Shadow Sovereign Ratings. ContemporaryEconomics, 9 (3), 367-384. http://dx.doi.org/10.5709/ce.1897-9254.175

Jiang, Y. (2018). Semiparametric Estimation of aCredit Rating Model. SSRN electronic journal. Avail-able at: https://www.garp.org//#!/risk-intelligence/all/all/a1Z1W000004B9vfUAC

Kamstra, M., Kennedy, P., & Suan, T.K. (2001). Combin-ing Bond Rating Forecasts Using Logit. Financial Review,36 (2), 75-96. https://doi.org/10.1111/j.1540-6288.2001.tb00011.x

Kaplan, R. S., & Urwitz, G. (1979). Statistical Models ofBond Ratings: A Methodological Inquiry. The Journal of Busi-ness, 52 (2), 231-261. http://dx.doi.org/10.1086/296045

Koulafetis, P. (2017). Credit Risk Transfer and Mitigation.In Modern Credit Risk Management (pp. 187-206). PalgraveMacmillan, London. ISBN 978-1-137-52407-2. https://link.springer.com/book/10.1057/978-1-137-52407-2

Longstaff, F., & Schwartz, E.S. (1995). “A Simple Ap-proach to Valuing Risky Fixed and Floating Rate Debt”.Journal of Finance, 50(3), 789-819. https://doi.org/10.1111/j.1540-6261.1995.tb04037.x

McConnell (2014). New Hedge Accounting ModelWill Improve Investor Understanding of Risk Manage-ment. Available at: https://www.ifrs.org/-/media/feature/resources-for/investors/investor-perspectives/investor-perspective-jun-2014.pdf

Merton, R.C. (1974). On the Pricing of Corporate Debt:The Risk Structure of Interest Rates. The Journal of Finance,29 (2), 449-470. https://doi.org/10.1111/j.1540-6261.1974.tb03058.x

Moody’s (2017). Annual Default Study: Corporate De-fault and Recovery Rates, 1920 - 2017. Available at: https://www.moodys.com/researchdocumentcontentpage.aspx?docid=PBC_1059749

Moody’s (2017b). Moody’s Financial Metrics KeyRatios by Rating and Industry for Global NonFin-ancial Corporates: December 2016. Available at:https://www.researchpool.com/download/?report/_id=1537315&show/_pdf/_data=true

Moody’s (2017c). Global Surface Transportation and Lo-gistics Companies. Moody’s Investor Service.

Moody’s (2017d). Equipment and Transportation RentalIndustry. Moody’s Investor Service.

Moody’s (2018). Annual Default Study: Corpor-ate Default and Recovery Rates, 1920 - 2017. Avail-able at: https://www.researchpool.com/download/?report/_id=1751185&show/_pdf/_data=true

Novotny-Farkas, Z. (2016). The Interaction of the IFRS9 Expected Loss Approach with Supervisory Rules and Im-plications for Financial Stability. Accounting in Europe,

http://www.bis.org/publ/bcbs161.pdf

https://www.jstor.org/stable/244122

https://pdfs.semanticscholar.org/b700/d412674ceae6030745437abe057894dc1cd6.pdf

https://pdfs.semanticscholar.org/b700/d412674ceae6030745437abe057894dc1cd6.pdf

http://www.jstor.org/stable/1831029

http://dx.doi.org/10.2139/ssrn.770805

http://dx.doi.org/10.2139/ssrn.770805

https://doi.org/10.3905/jsf.2018.24.1.039

https://doi.org/10.3905/jsf.2018.24.1.039

https://doi.org/10.1093/rof/8.4.537

https://doi.org/10.1093/rof/8.4.537

https://doi.org/10.1080/07350015.2014.902763

https://doi.org/10.1080/07350015.2014.902763

https://doi.org/10.1016/j.jeconom.2012.05.002

https://doi.org/10.1016/j.jeconom.2012.05.002

https://doi.org/10.1093/rfs/12.1.197

https://doi.org/10.1093/rfs/12.1.197



https://www.ey.com/Publication/vwLUAssets/Applying/_IFRS/_%E2%80%93/_IFRS/_9/_for/_non-financial/_entities//$FILE/Applying-FI-Mar2016.pdf




https://www.ey.com/Publication/vwLUAssets/ey-ifrs-9-expected-credit-loss//$File/ey-ifrs-9-expected-credit-loss.pdf



https://doi.org/10.1007/978-3-319-89494-2

https://doi.org/10.1007/978-3-319-89494-2

http://www.ifrs.org/News/Press-Releases/Documents/FCAGReportJuly2009.pdf

http://www.ifrs.org/News/Press-Releases/Documents/FCAGReportJuly2009.pdf

http://www.financialstabilityboard.org/wp-content/uploads/r/_0904a.pdf

http://www.financialstabilityboard.org/wp-content/uploads/r/_0904a.pdf

https://www.imf.org/external/np/sec/pr/2009/pdf/g20/_040209.pdf

https://www.imf.org/external/np/sec/pr/2009/pdf/g20/_040209.pdf

https://doi.org/10.1016/j.jbankfin.2004.06.010

https://doi.org/10.1016/j.jbankfin.2004.06.010

http://dx.doi.org/10.5709/ce.1897-9254.175

http://dx.doi.org/10.5709/ce.1897-9254.175

https://www.garp.org//#!/risk-intelligence/all/all/a1Z1W000004B9vfUAC

https://www.garp.org//#!/risk-intelligence/all/all/a1Z1W000004B9vfUAC



http://dx.doi.org/10.1086/296045

https://link.springer.com/book/10.1057/978-1-137-52407-2

https://link.springer.com/book/10.1057/978-1-137-52407-2



https://www.ifrs.org/-/media/feature/resources-for/investors/investor-perspectives/investor-perspective-jun-2014.pdf





https://www.moodys.com/researchdocumentcontentpage.aspx

https://www.moodys.com/researchdocumentcontentpage.aspx

https://www.researchpool.com/download/?report/_id=1537315&show/_pdf/_data=true





13 (2), 197–227. https://doi.org/10.1080/17449480.2016.1210180

Ohlson, J.A., (1980). Financial ratios and the probabilisticprediction of bankruptcy. J. Account. Res. 18 (1), 109–131.http://hdl.handle.net/10.2307/2490395

Ou, S., Chiu, D., & Metz, A. (2016). Annual DefaultStudy: Corporate Default and Recovery Rates, 1920-2015.Moody’s Investor Service, (May), 1–76. Available at:https://www.moodys.com/researchdocumentcontentpage.aspx?docid=PBC/_1018455

Schönbucher, P.J. (2003). Credit Derivatives Pricing Mod-els. 1st Ed. Chichester: John Wiley & Sons.

Tascón, M. T., & Castaño, F. J. (2012). Variables y modelospara la identificación y predicción del fracaso empresarial:revisión de la investigación empírica reciente (Variables andModels for the Identification and Prediction of Business Fail-ure: Revision of Recent Empirical Research Advances). Rev-ista de Contabilidad/Spanish Accounting Review, 15, 7-58.

Tsay, R.S., & Zhu, H. (2017). Market-Based Credit Ratingand Its Applications. Chapter 7 of Applied Quantitative Fin-ance. Springer. Germany. ISBN 978-3-662-54485-3.

Vazza, D., & Kraemer, N. (2018). 2017 AnnualGlobal Corporate Default Study and Rating Trans-itions. Standard and Poor’s report. Available at:https://www.spratings.com/documents/20184/774196/2017+Annual+Global+Corporate+Default+Study/a4cffa07-e7ca-4054-9e5d-b52a627d8639

https://doi.org/10.1080/17449480.2016.1210180

https://doi.org/10.1080/17449480.2016.1210180

http://hdl.handle.net/10.2307/2490395

https://www.moodys.com/researchdocumentcontentpage.aspx?docid=PBC/_1018455

https://www.moodys.com/researchdocumentcontentpage.aspx?docid=PBC/_1018455

https://www.spratings.com/documents/20184/774196/2017+Annual+Global+Corporate+Default+Study/a4cffa07-e7ca-4054-9e5d-b52a627d8639

